Forty-first KGS Computer Go Tournament

Sunday August 17th 2008

These results also appear on official KGS pages: Formal Division, Open Division which link to the game records.

Rules

 Formal divisionOpen division
format12-round Swiss8-round Swiss
board size9×913×13
rulesChineseChinese
komi
time8 minutes absolute13 minutes absolute

Times

The first round started at 08:00 UTC for the Formal and 08:05 for the Open division.

Results

As usual, the tournament was held in two divisions, Formal and Open, with more restrictive entry conditions for the Formal division.

Formal Division   9×9

placenamewinsSOSSoDOS
1stvalkyria116560
2ndCrazyStone97037
3rdFuego967620
4thAyaMC257111
5thMonteGNU56816
6thHouseBot0820

Open Division   13×13

placenamewinsSOSSoDOS
1stStoneCrazy63424
2ndFirstGoBot62614
3rdStoneGrid53214
4thAyaMC253014
5thGNU2330
6thHbotSVN0370

We welcomed Fuego, a new player to these events. It is an open-source program by Markus Enzenberger, Martin Müller, and Anders Kierulf, with contributions from David Silver and Niu, Xiaozhen. It played as Fuego9.

Many Faces of Go intended to play, but its registration email was not received until next day.

I was generally impressed by the high level of play in this tournament. I am 2-kyu: sometimes in previous events I have been impressed by individual moves, and thought "that was clever, I would not have thought of that". But in this tournament, there were whole games in which the players appeared to have a better understanding of the game than I have.

I assume that many of the improvements in strength we are seeing in these events are due to programmers finding better ways of making the most of MC/UCT and RAVE. It is good that after two years, these improvements are showing no signs of slowing.

Formal division

CrazyStone vs Fuego9
The position after move 22, marked.
In round 1, CrazyStone and Fuego9 played an interesting game SGF. They both played well, but CrazyStone played better. Its move 22, shown to the right, appeared to me to be decisive. After move 40 Fuego9 was reduced to thrashing around hopelessly. Some kibitzers were unimpressed by Fuego9 making moves that obviously did not work (when nothing worked), and by CrazyStone making moves that did not maximise its score (when it knew that they were enough to win).

In round 3, valkyria beat CrazyStone in a complicated and confusing game, in which many of the moves struck me as weird, without my actually being able to fault them SGF.

In round 6, valkyria played CrazyStone again, with the same colours as in round 3 SGF. It would be better if the draw program would reverse the colours when the same two programs meet for the second time. The games were the same for the first seven moves. Again valkria won.
      Valkyria appeared to me to be playing noticeably better than in previous events. Its author Magnus Persson told me "I have cut out inefficences in the search, and fine tuned the playouts. No big changes, but I feel a difference when I play it myself".

AyaMC vs Fuego9
The position after move 68, marked.
Also in round 6, AyaMC played the move shown to the left, against Fuego9. SGF. When I saw this move, I thought "this is hopeless for White, his left group only has two liberties". But Black can never quite put the two-liberty group in atari, and the marked move ensures that he never will be able to. This is discussed here. What is particularly impressive is that White knows this, and has been working towards it for several moves, ignoring a ko threat at a9 and sacrificing stones at b1 and d1.

In round 8, valkyria lost its only game of the tournament, to MonteGNU. SGF. Valkyria started badly, allowing MonteGNU to make the "first to extend from a crosscut" shape with move 5.

In round 9, MonteGNU appeared to misread a ladder in its game against CrazyStone. SGF. But maybe it knew it had no way to win.

The round 11 game between CrazyStyone and valkyria (with the colours reversed this time) was exciting. SGF. Valkyria started a ko fight that it couldn't win, losing eight stones in the bottom left in the knowledge that its ko threat in the top right, if unanswered, was enough to win it the game.

The round 12 game between valkyria and Fuego9 was another competent, if eccentric, game. SGF. These players, by using MC, act as if they "understand" kos, sacrifices, and sekis; things that classical programs have never excelled at, even though they are explicitly "explained", i.e. hard-coded, for them.

Open division

AyaMC2 failed to appear in time for round 1. Its opponent StoneGrid won by default. AyaMC2's owner Hiroshi Yamashita had forgotten to run it; he got it running in time for round 2.

In round 4, GNU killed all HBotSVN's stones before both players passed. They then disagreed about their status (in fact HBotSVN expressed no opinion, which KGS currently treats as a claim that they are all alive). During the clean-up, HBotSVN ran out of time.
       HBotSVN was playing well below its potential. Jason House informed me "It appears that I had too many versions of HouseBot running on my server. There were two instances running with kgsGtp name=HBotSVN and reconnect=t. They were competing with each other the whole tournament. The additional cpu and memory contention led to significantly reduced performance for the HouseBot login.". He later found that there was also a third copy of it running from the same server.

In round 5, against FirstGoBot, HBotSVN played absurdly many stones on the 1-line. SGF This may have been caused by the contention between the two copies of it mentioned above, including the way they kept reconnecting and thereby removing the other from the server. You can see their (different, alternating) log-in messages in the game record.

In round 8, AyaMC2 achieved a win against StoneCrazy. SGF. Aya has always been a strong program, but it has become noticeably stronger in recent months. If it had appeared in time for round 1, and if it had beaten StoneGrid in that round, it might have won the division.

 

Processor numbers, power, etc.

AyaMC and AyaMC2
Aya, running on AthlonX2 BE-2400 2.2GHz (2 cores per CPU, so both cores) and on Xeon X5355 2.66GHz 2 CPU (4 cores per CPU, so all 8 cores) repectively.
CrazyStone
Crazy Stone, running on AMD 4x2.2GHz
FirstGoBot
running on an Intel Core 2 Quad Q6600 at 2.4Ghz.
Fuego9
2x4 core, 2.5 GHz, 8GB
HBotSVN
Intel(R) Core(TM)2 Duo CPU T5450 @ 1.66GHz.   In fact due to an oversight there were three copies of it running on this system, competing for resources.
HouseBot
Intel(R) Core(TM)2 Duo CPU T5450 @ 1.66GHz
StoneCrazy
Crazy Stone, running on AMD 4x2.2GHz
StoneGrid
StoneGrid, running on an Intel Core 2 Duo L7700 1.8GHz.
valkyria
running on a single processor Pentium M, 1.4 Ghz.